e d

, Li

for variation in ozone concentrations of as much as 17 parts per billion. We conclude that residential ozone concentrations may

of upwind ambient NOx, reactive organic gases, and

atmospheric photochemistry, and models have been

2002). However, a recent critique of ozone mapping

efforts noted that interpolations from central site moni-

tors to local neighborhoods are not justified by the

Science of the Total Environment 3be over- or underestimated by measurements at a community monitor, depending on the variation in local traffic in the

community. These findings may have implications for studies of health effects of traffic-related pollutants.

D 2005 Elsevier B.V. All rights reserved.

Keywords: Ozone; Oxides of nitrogen; Vehicle emissions; Photochemistry

1. Introduction

There is an extensive literature describing how

ozone varies on a regional scale as a function of sources

developed to predict downwind ozone concentrations

in communities without monitoring stations (National

Research Council Committee on Tropospheric Ozone

Formation and Measurement, 1991; Diem and Comrie,Edward Avol a, John M. Peters a

aDepartment of Preventive Medicine, Keck School of Medicine at the University of Southern California, 1540 Alcazar Street,

CHP 236, Los Angeles, CA 90089, United StatesbSonoma Technology, Incorporated, 1360 Redwood Way, Suite C, Petaluma, CA 94954, United States

Received 24 June 2004; accepted 16 June 2005

Available online 10 August 2005

Abstract

Oxides of nitrogen in fresh traffic exhaust are known to scavenge ambient ozone. However, there has only been little study

of local variation in ozone resulting from variation in vehicular traffic patterns within communities. Homes of 78 children were

selected from a sample of participants in 3 communities in the southern California Childrens Health Study. Twenty-four hour

ozone measurements were made simultaneously at a home and at a community central site monitor on two occasions between

February and November 1994. Homes were geo-coded, and local residential nitrogen oxides (NOx) above regional background

due to nearby traffic at each participants home were estimated using a line source dispersion model. Measured home ozone

declined in a predictable manner as modeled residential NOx increased. NOx modeled from local traffic near homes accountedPredicting residential ozon

Rob McConnell a,*, Kiros Berhane a0048-9697/$ - see front matter D 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.scitotenv.2005.06.028

* Correspondin

3272.

E-mail address: rmcconne@usc.edu (R. McConnell).eficits from nearby traffic

ng Yao a, Frederick W. Lurmann b,

63 (2006) 166174

www.elsevier.com/locate/scitotenvand have not beenspatial resolution of available datag author. Tel.: +1 323 442 1096; fax: +1 323 442validated against the dense network of measurements

that would be required to justify the assumption that the

available in close proximity in urban areas, levels have

differed by up to 50% within 5 km of each other

Total(McNair et al., 1996).

One reason for the spatial inhomogeneity of ozone

is local variation in traffic, because nitric oxide (NO)

present in fresh vehicle exhaust reacts with and con-

sumes ozone. This reaction occurs much more rapidly

than the atmospheric photo-oxidation that produces

ozone regionally. Thus, ozone is reduced in tunnels, in

heavy traffic, and near heavy traffic corridors, com-

pared with nearby fixed site measurements near the

traffic corridor (Rodes and Holland, 1981; Chan et al.,

1991). Ozone concentrations are generally lower in

urban cores with heavy traffic, compared with down-

wind suburban areas with little traffic related NO to

react with and consume ozone (National Research

Council Committee on Tropospheric Ozone Forma-

tion and Measurement, 1991; Diem and Comrie,

2002; Gregg et al., 2003). In one previous study,

concentrations 8 m downwind from a freeway were

often less than 10% of background ambient levels,

and reduction in ozone concentrations were evident to

500 m (Rodes and Holland, 1981). Although this local

variation in traffic is known to modify ozone concen-

tration (Liu et al., 1997; Monn, 2001), there has been

little attempt to exploit complex variation in traffic on

freeways, arterials, and collector streets within neigh-

borhoods to predict ozone exposure at homes, predic-

tions which would be useful for health studies. Grid-

based photochemical air quality models typically have

K-theory dispersion algorithms that are not suitable

for applications with the fine horizontal resolution

(e.g., b200 m) needed to simulate line-source impact(Seinfeld and Pandis, 1998).

We investigated whether local variation in mea-

sured residential ozone in southern California could

be predicted based on traffic related oxides of nitrogen

(NOx) at the home. NOx from local traffic were

estimated from traffic counts available from the Cali-

fornia Department of Transportation.

2. Data and methodsdistribution of ozone is homogeneous within commu-

nities (Diem, 2003). Where monitoring sites have been

R. McConnell et al. / Science of theWe used an existing data set of ambient ozone

measured at homes in southern California and ofconcurrent measurements of ambient ozone and NOxat community central site monitors (Avol et al., 1998).

Residential NOx concentrations outside study homes

were estimated for our analysis from traffic patterns

near the homes, as described below. We examined the

relationship between home ozone and residential traf-

fic-related NOx, after adjusting for simultaneously

measured central site ozone and NOx. Based on this

regression, we estimated residential ozone at other

homes in the community.

Homes of 78 children were selected from a sample

of participants in the Childrens Health Study (Peters

et al., 1999). Details of the study population for this

residential ozone study and the sampling protocol

have been reported previously (Avol et al., 1998).

Briefly, ozone was measured between February and

December during 1994 immediately outside homes in

the southern California communities of Lancaster, San

Dimas, and Riverside. A controlled flow sampler,

which permitted timed exposure diffusion sampling

with an Ogawa sampler, was developed for use in this

study. Nitrite-coated filters from Ogawa USA (Pom-

pano Beach, FL) were loaded onto Teflon-filter

holders in an ozone-free environment. Before and

after sampling, filters were stored under conditions

designed to minimize ozone exposure. These passive

samplers were placed on the rear patio or porch at

least 1 m above the ground, but avoiding tree cano-

pies, roof overhangs, and home air vents. Sampling

was conducted during 24 h at each study home on two

occasions, once in the spring and once in late summer/

early fall. In previous evaluations (Koutrakis et al.,

1993) and in this study in co-located field sampling at

5 homes and in laboratory comparisons with a con-

tinuous ultraviolet photometer (Dasibi Model 1003-

AH, Glendale, California) (Lurmann et al., 1994; Avol

et al., 1998), the timed exposure diffusion sampling

results were found to be comparable to within

approximately 6%.

During each 24-hour home ozone measurement,

background ambient ozone was interpolated to the

neighborhood of each study home from measurements

at nearby air monitoring stations, typically located

within a few kilometers of study homes. These mea-

surements were made for comparison to each home

measurement. Ambient background NOxx in each

Environment 363 (2006) 166174 167community was measured at the nearest station.

These measurements were made with ultraviolet

Totaozone photometers and chemiluminescent NOx moni-

tors operated under prescribed regulatory air monitor-

ing protocols and procedures.

NOx levels above ambient background were esti-

mated from traffic patterns in the vicinity of each

home. Each address was standardized to United States

Postal Service specifications, using Lorton Data

(www.lortondata.com) and geo-coded using the Tele-

Atlas online geocoding service. Average annual daily

traffic volumes on freeways, arterials, and collectors

in each community were obtained from the California

Department of Transportation for 2000 and adjusted

to reflect 1994 traffic volumes, based on 3% per year

growth in Southern California (CalTrans, 2002). The

roadway locations were also based on the Tele-Atlas

roadway database. Fleet average vehicle emission

factors for NOx in 1994 were obtained from the

California Air Resources Boards EMFAC2002

model (California Air Resources Board, 2002). The

ambient concentrations of NOx from on-road vehicle

emissions on roadways within each community were

estimated at each residence using the CALINE4 line

source dispersion model developed by the California

Department of Transportation (Benson, 1989). The

modeling region for each community was a 20 km

by 20 km region encompassing the residences

wherein the CALINE4 model estimated unique con-

centrations at each geocoded residence location with

1020 m spatial resolution. These concentrations

should be regarded as incremental NOx contributions

that result in local variation on top of background

concentrations of ambient NOx (measured at the cen-

tral site in each community). We refer to these mod-

eled estimates as residential NOx to distinguish them

from measured central site NOx measurements. Back-

ground ambient NOx results from regional transport

and photochemistry on an urban and regional scale.

Thus, local traffic influences the relatively homoge-

neous background regional NOx concentrations down-

wind from these local traffic sources and also

influences middle-scale and neighborhood-scale var-

iations immediately adjacent to roadways and in the

nearby neighborhood. We elected to use the modeled

concentration of NOx for this study, because NO in

fresh traffic exhaust is rapidly converted to NO2, as it

mixes and reacts with background ozone,, and the

R. McConnell et al. / Science of the168modeled incremental contribution of traffic to NOxconcentration disperses rapidly from the source. Thus,modeled incremental NOx is expected to reflect NO in

fresh traffic exhaust available to react with ozone, and

if so, should predict decreases in ozone within com-

munities relative to the central site monitor.

Line source dispersion models like CALINE4 have

primarily been evaluated for short-term (1 h) esti-

mates of carbon monoxide concentration, where the

CALINE4 model has shown it is able to predict 1-h

concentrations within a factor of two (50 to +100%)in 80% of cases (Benson, 1989). We have evaluated

the CALINE4 models performance for long-term

NOx from local traffic at 12 southern California loca-

tions (see Fig. 1). The models annual predictions

were highly correlated (R2=0.97) with the 4-year

average concentrations observed at routine air mon-

itoring sites, yet the model underestimated the

observed concentrations since they included the regio-

nal background NOx. The CALINE4 model estimates

suggest local traffic contributions constitute 18% of

NOx concentrations in these communities.

An additional community in the original study,

Lake Gregory, was excluded from the analysis

reported here, because there was little traffic (Avol

et al., 1998). Seven other homes were excluded from

the analysis, because there were incomplete ambient

central site NOx measurements during the day of

sampling. In addition, in the original study, indoor

ozone was measured at homes, using a sampler iden-

tical to that described above. Two homes were

excluded because indoor ozone was unexplainably

greater than outdoor ozone. (Because there are gen-

erally no indoor sources of ozone, and ozone is sca-

venged indoors, indoor levels of ozone are virtually

always values lower than outdoor values). One other

outlier was excluded from the analysis, because out-

door home ozone was more than 3 standard deviations

from the mean and was inexplicably high with respect

to the central site monitor. Seventy-eight homes were

included in the final analysis.

Because there were two measurements at each

home, we needed to account for the correlation

between the repeated measures from the same subject.

We used the Generalized Estimating Equations (GEE)

approach (Diggle et al., 1994) to investigate the

effects of estimated residential NOx from the

CALINE4 model on outdoor residential ozone.

l Environment 363 (2006) 166174Under this approach, the interpretation of parameter

estimates is identical to that in standard multiple linear

once

d NO

liforn

Totalregression, but it has the advantage of providing

correct standard errors (which would have been

underestimated otherwise). The assumptions for the

GEE based model are similar to those in standard

0 20 40Observed Ambient NOx C

CALI

NE4

mod

eled

NO

x fro

m lo

cal t

raffi

c (p

pb)

0

2

4

6

8

10

12

14

Fig. 1. Comparison of annual average CALINE4 model estimate

concentrations observed at 12 air monitoring stations in southern Ca

R. McConnell et al. / Science of themultiple linear regression. The following final model

was fitted:

Ycij a dTZcij c1CS Ozonecij c2CS NOx cij c3NOx cij ecij: 1

Where Y is ozone (measured at home), c, i and j

denote community, individual home, and visit, respec-

tively. Here, a represents the overall intercept and drepresents a vector of parameters of effects of a matrix

of adjustment variables, Zcij (e.g., visit, community).

The parameter of main interest was c3, the effect ofresidential NOx levels, NOx cij, in a model that

adjusted for central site O3 and NOx levels, denoted

by CS_Ozonecij and CS_NOx cij, respectively. Central

site ozone and NOx measurements were centered at

the overall mean of 35.9 parts per billion (ppb) and

44.6 ppb, respectively, in these models. This GEE

model was fitted with the identity link and a

bworkingQ exchangeable correlation structure. It usedbempiricalQ standard errors that are robust to any mis-specification of the nature of dependence between the

two repeated measures from the same home.We assessed how generalizeable the estimates for

the effect of residential NOx might be to other com-

munities by evaluating the effect of NOx in models

without the community indicator variables. We eval-

60 80ntration (ppb)

Atascadero

Santa Maria

Lompoc

Lancaster

Lk Arrowhead

San Dimas

Glendora

Upland

Mira Loma

Riverside

Lk Elsinore

Alpine

y = 0.18x - 0.5,R2 = 0.97

x concentrations from local traffic to 19951998 average NOxia.

Environment 363 (2006) 166174 169uated potential modification of the effect of residential

NOx on residential ozone by community and by visit

early or late in the year. Appropriate interaction terms

were added to the GEE model to test these hypoth-

eses. A validation of the model consisted of randomly

selecting 2/3 of the sample in each community and

fitting a model which was then used to predict the

observed O3 in the remaining 1/3 of the sample.

Finally, the predictions from the model were used

to examine the possible impact of residential NOx on

long-term residential ozone concentrations at homes

of all participants in the Childrens Health Study in

these 3 communities with addresses that could be

accurately geo-coded (449 in Lancaster, 498 in River-

side, and 439 in San Dimas). For this prediction, we

used 1994 annual average residential NOx concentra-

tions derived from CALINE4 at the GIS coded

addresses of all study homes and the 19941998

average ozone and NOx concentrations measured at

the central site monitors. We also estimated the effect

based on a single 1994-year average for the measured

central site pollutants. Because the home ozone pre-

dictions were very similar, we have presented only the

estimates based on the more stable multi-year pollu-

tant average.

3. Results

Average 24-h ozone measurements were 33 ppb

and 34 ppb at the homes and central site monitoring

stations, respectively, but there was a large range from

4.5 ppb to over 90 ppb (Table 1). Estimated residential

NOx concentrations modeled from traffic were not

normally distributed. The median was 11 ppb, but

asured at central sites, and of modeled residential NOx

mes) San Dimas (N =22 homes) All communities (N =78 homes)

max) Mean (SD) (min, max) Mean (SD) (min, max)

79) 28 (15) (4.5, 60) 33 (15) (4.5, 79)

92) 34 (14) (15, 57) 36 (15) (9.7, 92)

42) 56 (21) (26, 103) 45 (30) (3.2, 142)

Median (min, max) Median (min, max) Median

IQR)

(min, max) Median

(IQR)

(min, max)

35) 16.2 (3.1) (6.2, 24) 11 (9.7) (2.1, 35)

Table 2

Determinants of home ozone concentration

Parameter Coefficient (in pbb) (S.E.) P value

Residential NOxa 0.51 0.11 b0.0001

Central site O3a 0.91 0.05 b0.0001

Central site NOxa 0.029 0.018 0.10

Riverside 0.28 1.65 0.86San Dimas 0.12 2.03 0.95

Visit 1 2.29 1.0 0.03

Intercept 37.7 1.14 b0.0001a Residential NOx in ppb; central site O3 and NOx centered at 40

and 45 ppb, respectively.

R. McConnell et al. / Science of the Total Environment 363 (2006) 166174170(IQR) (IQR)

Residential NOxb 4.8 (1.5) (2.1, 6.8) 11 (5.8) (6.2,one home 90 m from a freeway had an estimated

NOx exposure of 35 ppb. Fifty percent of the esti-

mated concentrations were between 6.2 and 16 ppb;

and 80% were between 4.3 and 19 ppb (results not

tabulated). Lancaster had a much smaller range of

modeled NOx than either of the other two commu-

nities. Although these exposures are modest compared

with the much larger measured central site NOx mea-

surements, it is important to note that these were

incremental concentrations from traffic near homes

above any community ambient background levels.

Residential ozone was inversely associated with

estimated residential NOx, decreasing by 0.51 ppb

for each modeled increase of 1 ppb in residential

NOx (Table 2). Thus, the range of residential NOx in

this study (2.135 ppb from Table 1) was associated

with almost a 17 ppb decrease in the ozone measured

at those homes (results not tabulated). However, a

relatively small proportion of homes near heavy traffic

corridors accounted for most of the variation in ozone,

and the variation due to traffic at most homes was

modest. The 10th to the 90th percentile of the resi-

Table 1

Concentrations of ozone measured at homes, of ozone and NOx me

Pollutant Lancaster (N =25 homes) Riverside (N =31 ho

Mean (SD) (min, max) Mean (SD) (min,

Home O3a 40 (60) (6.5, 63) 32 (14) (9.6,

Central site O3a 39 (13) (16, 65) 36 (16) (9.7,

Central site NOxa 29 (16) (3.2, 60) 46 (37) (11, 1a Measured 24-h mean (ppb).b Modeled for annual average daily traffic counts.dential NOx distribution (from 4.3 to 19 ppb) was

associated with a 7.5 ppb decrease in home ozone.

Home ozone exposure was higher during the first visit

(earlier in the year), on average, than on the second

visit, after adjusting for central site ozone measure-

ments and other variables in the model (Table 2). Visit

was also significant in a model adjusting for season

(before May or after October), but the addition of

season to the model did not substantially change the

effect of NOx (data not presented). We adjusted for

central site NOx, an indication of more polluted days

with stagnant air around roadways, in order to account

for potentially larger effects of residential traffic on

ozone on those days.

The split sample resulted in coefficients derived

from 2/3 of the homes similar to those from the entire

sample. The coefficient for residential NOx in 2/3 of the

sample was0.51, for ambient ozone at the central site0.89, for visit 2.26. Although the power to test this

prediction model in the remaining homes was limited

by small sample size, the predicted home ozone in the

remaining homes was highly correlated with the mea-

A prediction curve for home ozone across the

range of incremental traffic-modeled NOx observed

in this study, in a hypothetical community with mea-

sured long-term central site monitoring station aver-

age NOx, CS_NOx, and ozone, CS_Ozone, has the

following form:

Home ozone 37:7 0:919CS Ozone 35:90:029 CS NOx 44:6 0:51NOx

2where long-term (19941998) central site ozone and

NOx measurements were centered at the overall sam-

Total Environment 363 (2006) 166174 171sured concentration (R =0.94; p b0.0001). However,this was largely accounted for by the large variability

in ozone at the central site, which was highly correlated

with home ozone. The residuals from the prediction

model excluding residential NOx were more modestly

but still significantly correlated with residential NOx(R =0.43; p =0.002).

We considered whether the prediction model

derived from the entire sample could be used to pre-

dict ozone at homes in other communities in southern

California, based on ambient measurements at a com-

munity central site monitor and traffic patterns near

homes, exposure indices that are widely available

throughout California (and elsewhere in the United

States). Our premise was that the atmospheric chemi-

cal reactions affecting these relationships and the

general nature of vehicle emissions not accounted

for in the vehicle emission models (e.g., NO/NOxratios) would be similar across many communities.

However, this type of statistical modeldescribing the

physical effects of traffic-related NOx on local ozone

could not be extrapolated to other communities if

residential ozone or NOx were confounded by com-

munity, or if the effect of residential NOx were to vary

in different communities, after adjusting for central

site measurements and other covariates in the model.

None of these conditions occurred in our limited

sample of communities: There were no significant

effects of community (Table 2); the coefficient for

residential NOx was almost identical in a model that

did not adjust for community (data not shown); and

there were no significant interactions between resi-

dential NOx and community (that is, the effect of

residential NOx did not differ by study community).

The significant effect of visit, an unphysical parameter

required in the model, suggests that predicted residen-

tial ozone with respect to central site measurements

may vary by time of the year or other unmeasured

conditions prevalent at the different visits in this

study. However, any bias introduced by this effect

was unlikely to vary by residential NOx, because

there was no interaction of residential NOx with

visit (data not presented). Although this consideration

may make the accuracy of ozone predictions ques-

tionable and merits further investigation, it is never-

theless instructive to examine the likely effect of

R. McConnell et al. / Science of thetraffic on home ozone concentrations with respect to

long-term central site ozone concentrations.ple mean in our study of 35.9 ppb and 44.6 ppb,

respectively. Traffic modeled residential NOx were

estimated from CALINE4 at homes. Visit and com-

munity (with visit 2 and Lancaster as references,

respectively) should not affect the long-term within-

community variability in ozone with respect to the

central site, nor should they contribute to the varia-

bility in the predicted values with respect to other

homes in the community. Based on this prediction

equation, we estimated the distribution of residential

ozone in the entire cohort of children enrolled in the

Childrens Health Study in the three communities. We

restricted the distribution to homes with residential

NOx values less than 35 ppb, because the behavior

of the model at higher residential NOx (which were as

large as 128 ppb in this population) is unknown. In

San Dimas and Riverside the predicted range in resi-

dential ozone was 15 ppb (Table 3), although the

variation from the 10th to the 90th percentile of the

distribution in those towns was a more modest 4.5 in

Table 3

Predicted 5-year average home ozone (in pbb) in entire cohort, by

community

Town Predicted 5-year O3 at homes NOxa

(min,

max)

O3 at

C.S.aMean 10%,

90%aMin,

max

Lancaster

(N =449)

33 32, 34 29, 35 0.3, 12 33

Riverside

(N =498)

27 23, 29 16, 31 4.1, 35 31

San Dimas

(N =439)

17 14, 19 9, 24 2.8, 33 23

a Ozone 10%, 90% corresponds to lower decile and upper decileof the distribution; NOx (in ppb) corresponds to residential NOx; O3at C.S. was measured at the central site monitor in each community.

It is well known that measurements of regional

pollutants like ozone at central site monitors misclas-

Totasify personal exposure (Delfino et al., 1996; United

States Environmental Protection Agency, 1996). The

estimates in Table 3 suggest that there is likely to be

systematic spatial variation of average ozone with

nearby traffic. The predicted within-communitySan Dimas and 5.8 in Riverside. However, because

we truncated the upper distribution of residential NOxat 35 ppb, lower ozone levels (with a lower limit of

zero) might be expected to be observed at homes with

higher levels of residential NOx in these communities.

4. Discussion

There has been little previous research demonstrat-

ing the predictable local spatial variability in ozone

that we have shown in these study communities as

nearby traffic varies. As described above, ozone con-

centrations at most homes varied by a modest 7.5 ppb

or less due to nearby traffic. However, reductions in

ozone concentrations of up to 17 ppb were associated

with heavy local traffic across the range of estimated

residential NOx in this sample. Our results are con-

sistent with one previous report that demonstrated a

3040% decrement in ozone as traffic and building

density increased at many sites in a neighborhood

within 2 km of central monitor (Lin et al., 2001).

There are several potential implications of this obser-

vation for the evaluation of health effects of traffic and

ozone, and potentially for regulatory policy: (1) The

interpretation of measurements of ambient ozone at

community monitoring stations in southern California

and elsewhere may be biased if traffic patterns at

homes in the community differ substantially from

traffic near the monitoring station; (2) there should

be further investigation of ozone variability due to

traffic on a neighborhood scale within communities in

order to refine models that could be used for assigning

ambient ozone exposures at homes for use in epide-

miologic studies; and (3) the inverse relationship of

ozone to traffic may bias the results of ongoing studies

of the health effects of fresh traffic emissions, if

potentially confounding health effects of ozone are

not appropriately accounted for.

R. McConnell et al. / Science of the172range of 15 ppb in measurement of home ozone

concentration is small compared with the large tem-poral variability shown in Table 1, and the levels are

low compared with the current National Ambient Air

Quality Standards regulating peak concentrations

exceeding 120 ppb for 1 h and 80 ppb for 8 h of

the day. However, the long-term average 24 h ozone

measured at the central sites was relatively low, even

in these study communities with high peak ozone

concentration, which were selected to represent

diverse southern California ozone profiles. Therefore,

the predicted misclassification of ozone exposures at

homes could constitute a substantial proportion of the

measured long-term ozone exposures at central sites.

Because NO from traffic near a central site monitoring

station would also reduce the measured ozone at that

station, the central site measurements could either

under-estimate or over-estimate the average commu-

nity home ozone exposure, depending on whether

nearby traffic at the central site were substantially

higher or lower than near homes in the surrounding

community. The U.S. Environmental Protection

Agency has recommended that regional air pollution

monitoring stations be sited away from roads with

heavy traffic, and previous work in Los Angeles has

demonstrated that ozone concentrations are affected

up to at least 500 m of a major roadway (Rodes and

Holland, 1981). However, using CALINE4 we have

estimated NOx related to long-term average local

traffic to contribute as much as 38 ppb at the central

site monitoring station (in Long Beach) used for

characterizing exposure in a multi-year longitudinal

health study (Peters et al., 1999). Therefore, local

traffic around such stations is potentially an important

source of error in assigning exposure to an entire

community based directly on central-site monitoring

station data.

The model presented illustrates how ozone might

be predicted at individual homes. There was substan-

tial variability in the ozone measurements at homes

and central sites, and there was a large range of nearby

traffic at these homes, representative of many com-

munites in southern California and likely in other

regions of the United States. However, questions

remain regarding this approach that cannot be

answered with the available data, and additional mea-

surements over a longer time frame would be useful to

model better the long-term variation in ozone expo-

l Environment 363 (2006) 166174sure at homes. The 24 h ozone varied markedly over

the period of the study, so the effect of residential NOx

effect of residential NOx on ozone is unlikely

to vary by visit. Therefore, to the extent that

Totalexplained a relatively small proportion of the varia-

bility in residential ozone. A larger proportion of

variability in the longer-term average residential

ozone might be explained by residential NOx. Resi-

dential NOx also explained only a modest proportion

of measured home ozone based on the split sample

approach (R =0.43 for NOx with the residual afteraccounting for the effect of central site ozone),

although this may be due in part to the small sample

size. Also, although the range of variability in resi-

dential NOx was large, there were relatively few

homes that accounted for the higher residential NOxestimates. It would be useful to examine the effect of

residential NOx on ozone at more homes with heavy

local traffic to evaluate the stability of observed esti-

mates in areas where more marked reductions in

ozone might be expected. Additional measurements

in these communities also would help determine the

relevance of local traffic patterns for the higher ozone

concentrations observed between 10 am and 6 pm

(when children and other susceptible groups may

engage in more outside activity that would increase

exposure).

We observed no difference in the effect of resi-

dential NOx in different communities, using a model

that considered only linear effects of residential NOx(because this model is consistent with the known

chemistry of ozone and local sources of NOx on the

neighborhood scale of interest). However, the statis-

tical power to identify interactions was limited. In a

sensitivity analysis that used flexible techniqures

that relaxed the assumption of linearity (Hastie

and Tibshiarni, 1990), we observed some non-line-

arity in the effect of NOx that may have reflected

differences in the effect in Lancaster, which had

much smaller residential NOx estimates than either

of the other two communities. Longer term home

ozone and NOx measurements would allow better

assessment of heterogeneity across communities. In

addition, there was systematic variability in ozone

associated with the measurement at the second visit

to the homes, after adjusting for other covariates

(Table 2). Because traffic-modeled NOx estimates

using CALINE4 were based on average traffic

count estimates and on average wind speed and

direction over an entire year, meteorological condi-

R. McConnell et al. / Science of thetions or traffic patterns that differed on different

days of measurement from those used to estimatevisit represented unmeasured influences on ozone

throughout the community, including at the central

site, there may be little bias to the assessment of

how a home varies with respect to a central site

within a given community, which is the key para-

meter of interest. This issue requires further study.

Finally, estimating long term average effects of

residential NOx on ozone would not be possible in

communities in which there is new development

with traffic patterns that change in different ways

in different parts of the community.

Although uncertainties in the model presented

require further assessment, the observed variation

in ozone may be important for the design and

interpretation of ongoing studies of traffic-related

pollution. Heavy local traffic has been associated

with deficits in lung function, with asthma and

with asthmatic symptoms in children (Brauer et

al., 2002; Delfino, 2002). It has been shown that

particle number and ultrafine particle mass decrease

dramatically to background levels within short dis-

tances downwind of freeways (Zhu et al., 2002),

and it is biologically plausible that such exposure

would cause the observed associations with respira-

tory health (Li et al., 2003). It is likely that expo-

sure to fine and ultrafine particulate matter and

other air pollutants in fresh traffic exhaust increases

not just with freeways, but also with nearby traffic

on arterial and collector roadways (although there

has been little study of this issue). Our results

indicate that exposure to these local toxic traffic

related pollutants is likely to be inversely correlated

with ozone exposure. Therefore, in high ozone com-

munities the assessment of toxic effects of traffic

likely also will be confounded by the toxic effects

of ozone.

5. Conclusion

This study suggests that traffic patterns near homesthe long-term average NOx might explain the effect

of visit. However, we observed no evidence for an

interaction between visit and residential NOx, so the

Environment 363 (2006) 166174 173are potentially a useful tool to predict local spatial

variation in ozone within communities.

R. McConnell et al. / Science of the Total Environment 363 (2006) 166174174Acknowledgements

Tami Funk and Siana Alcorn of Sonoma Technol-

ogy contributed to developing traffic-modeled expo-

sures and performed extensive quality assurance of

the air pollution data, respectively, used for this study.

Daniel Stram and W. James Gauderman made useful

suggestions on modeling strategies. Edward Rappa-

port, Jassy Molitor and Vince Lin provided program-

ming support. We acknowledge the hard work of

Helene Margolis and Dane Westerdahl of the Califor-

nia Air Resources Board and the staff at the partici-

pating air quality districts.

This studywas supported by the Southern California

Particle Center and Supersite, the National Institute of

Environmental Health Science (grants P30 ES07048,

P01 ES09581, and P01 ES11627), the Environmental

Protection Agency (grant R 82670801-3), the Califor-

nia Air Resources Board (Contract 94-331), and the

Hastings Foundation.

References

Avol EL, Navidi WC, Colome SD. Modeling ozone levels in and

around Southern California homes. Environ Sci Technol 1998;

32:4638.

Benson P, 1989. Caline4a dispersion model for predicting air

pollution concentration near roadways. Sacramento, CA7 Cali-fornia Department of Transportation, Office of Transportation

Laboratory; 1989. p. 205.

Brauer M, Hoek G, Van Vliet P, Meliefste K, Fischer PH, Wijga A,

et al. Air pollution from traffic and the development of respira-

tory infections and asthmatic and allergic symptoms in children.

Am J Respir Crit Care Med 2002;166(8):10928.

California Air Resources Board. EMFAC2002; California air

resources boards emissions inventory series, the latest update

to the on-road emissions inventory. Sacramento, CA7 CaliforniaAir Resources Board; 2002. Available at http://www.arb.ca.gov/

msei/onroad/onepagers/2002.pdf.

CalTrans. California motor vehicle stock, travel and fuel forecast.

Sacramento, CA7 California Department of Transportation;2002. Available at http://i80.dot.ca.gov/hq/tsip/TSIPPDF/

MVSTAFF02.pdf.

Chan C, Ozkaynak H, Spengler JD, Sheldon L. Driver exposure to

volatile organic compounds, CO, ozone and NO2 under different

driving conditions. Environ Sci Technol 1991;25:96472.

Delfino RJ. Epidemiologic evidence for asthma and exposure to air

toxics: linkages between occupational, indoor, and community

air pollution research. Environ Health Perspect 2002;110(Suppl.

4):57389.

Delfino RJ, Coate BD, Zeiger RS, Seltzer JM, Street DH, KoutrakisP. Daily asthma severity in relation to personal ozone exposureand outdoor fungal spores. Am J Respir Crit Care Med

1996;154(3 Pt 1):63341.

Diem JE. A critical examination of ozone mapping from a spatial-

scale perspective. Environ Pollut 2003;125:36983.

Diem JE, Comrie AC. Predictive mapping of air pollution involving

sparse spatial observations. Environ Pollut 2002;119(1):99117.

Diggle PJ, Liang KY, Zeger SL. Analysis of longitudinal data. New

York7 Oxford University Press; 1994.Gregg JW, Jones CG, Dawson TE. Urbanization effects on tree

growth in the vicinity of New York City. Nature 2003;

424(6945):1837.

Hastie TJ, Tibshiarni RJ. Generalized additive models. London7Chapman and Hall; 1990.

Koutrakis P, Wolfson J, et al. Measurement of ambient ozone using

a nitrite-coated filter. Anal Chem 1993;65:20914.

Li N, Sioutas C, Cho A, Schmitz D,Misra C, Sempf J, et al. Ultrafine

particulate pollutants induce oxidative stress and mitochondrial

damage. Environ Health Perspect 2003;111(4):45560.

Lin T-Y, Young L-H, Wang C-S. Spatial variations of ground level

ozone concentrations in areas of different scales. Atmos Environ

2001;35:5799807.

Liu LJ, Delfino R, Koutrakis P. Ozone exposure assessment in a

southern California community. Environ Health Perspect 1997;

105(1):5865.

Lurmann F, Roberts P, Main H, Hering S, Avol E, Colome S. Phase

II Report Appendix A: Exposure Assessment Methodology;

Final Report on Contract A033-186 to the California Air

Resources Board. Sacramento; 1994.

McNair LA, Harley RA, Russell AG. Spatial inhomogeneity in

pollutant concentrations, and their implications for air quality

model evaluation. Atmos Environ 1996;30:4291301.

Monn C. Exposure assessment of air pollutants: a review on spatial

heterogeneity and indoor/outdoor/personal exposure to sus-

pended particulate matter, nitrogen dioxide and ozone. Atmos

Environ 2001;35:132.

National Research Council Committee on Tropospheric Ozone

Formation and Measurement. Rethinking the ozone problem

in urban and regional air pollution. Washington, DC7 NationalAcademy Press; 1991.

Peters JM, Avol E, Navidi W, London SJ, Gauderman WJ, Lurmann

F, et al. A study of twelve Southern California communities with

differing levels and types of air pollution: I Prevalence of

respiratory morbidity. Am J Respir Crit Care Med 1999;

159(3):7607.

Rodes CE, Holland DM. Variations of NO, NOx, and O3 concen-

trations downwind of a Los Angeles freeway. Atmos Environ

1981;15:24250.

Seinfeld JH, Pandis SN. Atmospheric chemistry and physics: from

air pollution to global change. New York7 J. Wiley and Sons;1998.

United States Environmental Protection Agency, Air Quality Cri-

teria for Ozone and other Photochemical Oxidants. Research

Triangle Park, NC7 Office of Research and Development;1996.

Zhu Y, Hinds WC, Kim S, Sioutas C. Concentration and size

distribution of ultrafine particles near a major highway. J AirWaste Manage Assoc 2002;52(9):103242.

Predicting residential ozone deficits from nearby trafficIntroductionData and methodsResultsDiscussionConclusionAcknowledgementsReferences